CN102842066B - A kind of modeling method of biomass stove burning optimization - Google Patents

Abstract

The present invention relates to a kind of modeling method of biomass stove burning optimization.First the inventive method gathers the characteristic index of biomass stove operational factor and relevant characterising biological matter stove fired state, building database; Then use least square method supporting vector machine and radial base neural net for different fuel respectively, set up combustion model, determine the portfolio ratio of least square method supporting vector machine model and radial basis neural network, least square method supporting vector machine model is combined with the scale-up factor of radial basis neural network by the optimum determined, form built-up pattern, for other kind of biomass fuel modeling of given biomass stove, the burning optimization model group of different biomass fuels is combined, forms a biomass stove burning optimization block mold.The inventive method meets fuel in biomass stove burning optimization and changes and the actual requirement of the limited change of fuel type, ensure that accuracy and the feasibility of biomass stove burning optimization.

Description

A kind of modeling method of biomass stove burning optimization

Technical field

The invention belongs to information control technology field, relate to a kind of modeling method of biomass stove burning optimization.

Background technology

The method of biomass stove burning optimization is the important technical of energy-saving and emission-reduction, its target is under certain load (biomass fuel delivery rate) condition, is obtained the running status of high-level efficiency, low pollution emission by the operational factor of adjustment biomass stove air distribution.The collocation of the air distribution parameter of biomass stove has direct impact to biomass stove fired state, and the configuration of the operating parameters such as different air distributions, oxygen amount directly can cause different burning efficiency and the discharge capacity of dusty gas.For given biomass stove, under certain loading condiction, for different fired state characteristic indexs, there is a kind of air distribution scheme of optimum, the characteristic index optimization of corresponding fired state can be made, but, have complicated coupled relation between the operating parameter of biomass stove, the configuration of optimum operating parameter be found and be not easy.The modeling of biomass stove combustion characteristics is the key issue in biomass stove burning optimization, is not also well resolved at present.

In reality, the burning optimization of biomass stove mainly runs by staff's experience, and often just maintain the burning of biomass stove, the parameter configuration therefore in actual motion also exists larger room for promotion.

Summary of the invention

Target of the present invention is for the bottleneck problem in biomass stove burning optimization, proposes a kind of modeling method taking into account model prediction accuracy and generalization ability.

First the present invention specifically gathers the combustion case data of biomass stove, carry out the selection of modeling data and suitable pre-service, to ensure predictive ability and the generalization ability of model, the corresponding modeling algorithm of final application and optimized algorithm set up the combustion characteristics model of biomass stove.The method ensure that precision of prediction and the generalization ability of model by data selection and pre-service.

Technical scheme of the present invention is by the collection of biomass stove burning data, modeling, the selection of data sample and pre-service, set up the combustion characteristics model of biomass stove, a kind of modeling method of biomass stove burning optimization of establishing, utilize the method to set up comparatively accurately and the stronger biomass stove burning optimization characteristic model of generalization ability.

The step of the inventive method comprises:

Step (1). gather the characteristic index of biomass stove operational factor and relevant characterising biological matter stove fired state, building database; Concrete biomass stove operational factor is obtained by biomass stove real-time running data storehouse, or directly measures collection by instrument and equipment.General biomass stove right biomass fuel kind (originating different with technical analysis data) limited, therefore, different biomass fuels (raw material of biomass fuel is different with source) is planted for given biomass stove burning n (n >=1), different types of fuel will separate image data, so that modeling respectively targetedly.

Described biomass stove operational parameter data comprises: the technical analysis index of biomass fuel, primary air velocity, secondary wind speed, oxygen amount, after-flame wind speed, biomass fuel delivery rate; The data of the characteristic index of described characterising biological matter stove fired state comprise NOx concentration in flue gas and biomass stove burning efficiency, and its method obtained is mature technology;

Step (2). the data in database are selected and pre-service, and use least square method supporting vector machine and radial base neural net (RBF network) for different fuel respectively, combustion model between the characteristic index setting up biomass stove operation parameter and fired state, therefore, be directed to n kind biomass fuel, set up n model respectively.Concrete grammar is:

Because biomass stove fuel change is limited, therefore respectively for the combustion case of different fuel, carry out data selection in a database, select the sample data of modeling, following principle is followed: be 1. evenly distributed during selection, namely it is uniform for spatially distributing at the topological structure of the input quantity of model, and the input quantity of selected data is not intensive concentrate on a point, can occupy a space uniformly; 2. equal in number, in topological structure, be namely positioned at the sample size of the data of difference, should be more or less the same in 10 ﹪ of the sample data amount of minimum data point, the data volume of certain point can not be made so a lot, and the data volume of other points is little, to ensure modeling quality;

Pre-service before modeling is carried out to the data selected, by the conversion of unit or the method for multiplying factor, makes the data of each input quantity and corresponding output quantity be in the order of magnitude and differ the state being less than 1, then input quantity is normalized;

The data that application pre-service is good, first least square method supporting vector machine algorithm is adopted, for a kind of fuel modeling, least square method supporting vector machine algorithm institute established model generalization ability is stronger, and then apply radial base neural net modeling, radial base neural net institute established model empiric risk is less, finally least square method supporting vector machine model and radial basis neural network is carried out being combined to form the final burning optimization model being directed to a kind of fuel; Output parameter for the input parameter of modeling and the characteristic index of characterising biological matter stove fired state is expressed as , wherein represent the organize the biomass stove operational factor vector as input data, represent the group as the parameter of the characterising biological matter stove fired state feature of output parameter, for sample size, based on actual operating data, set up the model between operation parameter fired state index different from biomass stove;

First, adopt the modeling of least square method supporting vector machine algorithm, kernel function elects radial basis function as:

for the width of radial basis function, this representation is canonical form; for mapping function, if required objective function is: , for the characteristic index predicted value of the boiler combustion status that model exports, for weight coefficient vector, for intercept; Introduce relaxation factor ξ ^* _iand ξ _iand allow error of fitting ε, ξ ^* _i>=0, ξ _i>=0, model is by retraining:

, under condition, minimize:

Obtain, wherein constant cfor penalty coefficient, c>0; This minimization problem is a convex quadratic programming problem, introduces Lagrangian function:

Wherein , , , for Lagrange's multiplier, >=0, >=0, >=0, >=0.

At saddle point place, function L is about w, b, ξ _i, ξ _i ^*, be also , , , maximal point, minimization problem is converted into the maximization problems asking its dual problem;

LagrangianL is about w, b, ξ at saddle point place _i, ξ _i ^*minimal point:

The dual function of Lagrangian function can be obtained:

Now,

According to Ku En-Plutarch (KKT) conditional theorem, following formula is had to set up at saddle point:

From above formula, α _i α _i ^* =0, α _i with α _i ^* can not be all non-zero simultaneously, can obtain:

Can b be obtained from above formula, obtain model;

Secondly, the radial base neural net modeling that learning ability sum functions approximation capability is strong is adopted:

For its output of the radial base neural net of individual hidden node is : , for weight coefficient, for dimension input vector, be the center of individual basis function, for the base width parameter of function; The key setting up radial basis neural network is the center determining basis function , sound stage width degree and weight coefficient ; Adopt particle cluster algorithm repetitive exercise radial base neural net, definition particle cluster algorithm initial population each dimension component of vector, be respectively hidden node, number Basis Function Center, the sound stage width degree of function and weight coefficient, objective function is: , wherein be the radial base neural net output valve of individual sample, be the actual value of individual sample; When reach minimum, when reaching setting value or complete iterations, training completes, obtain hidden node number, Basis Function Center, the sound stage width degree of function and weight coefficient, thus obtain radial basis neural network;

Step (3). determine the portfolio ratio of least square method supporting vector machine model and radial basis neural network; For the same biomass fuel that modeling data is corresponding, gather the data of new biomass stove under different running status as test samples, application least square method supporting vector machine model and the average weighted Forecasting Methodology of radial basis neural network, predict check data, namely , wherein be the target prediction value of group test samples operating mode, for least square method supporting vector machine model predication value, for radial basis neural network predicted value, for least square method supporting vector machine model predication value scale-up factor, for the prediction scale-up factor of radial basis neural network, and ;

with determination adopt the optimizing of ant group algorithm iteration to determine, initialization ant group position vector each dimension component, be respectively supporting vector machine model ratio with original Model Weight , objective function is: , wherein be the error of the biomass stove combustion characteristic index that group operating mode real data and combination model are predicted, K is test samples quantity, when variance summation achieve minimum, when reaching setting value or complete iterations, optimizing completes;

Step (4). least square method supporting vector machine model is combined with the scale-up factor of radial basis neural network by step (3) determined optimum, forms built-up pattern, namely , wherein Z is the built-up pattern after upgrading, thus realizes a kind of foundation of burning optimization model of biomass stove fuel;

Step (5). for other kind of biomass fuel modeling of given biomass stove; Modeling procedure is consistent with step (2) ~ (4), obtains the combustion model in different biomass fuel situation, total n kind fuel, therefore correspondence establishment n model as described in step (1);

Step (6). the burning optimization model group of different n kind biomass fuels is combined, forms a biomass stove burning optimization block mold; When needing calling model to predict, based on fuel index, calls corresponding burning optimization model and predicts.

The present invention passes through data mining, in operational factors different in a large number combination, the method of applied for machines study, the relational model between the characteristic index excavating operational factor and biomass stove fired state, then be very potential method in conjunction with the burning optimization that optimized algorithm carries out biomass stove.How to make the method really reach biomass stove burning and produce actual requirement, be a puzzlement engineering technical personnel's difficult problem, main bugbear is, how to improve prediction and the generalization ability of model, makes it to reach the effect of the more accurate burning optimization taking into account indices.

The modeling method that the present invention proposes effectively can carry precision of prediction and the generalization ability of biomass stove combustion model, and fuel in biomass stove burning optimization can be met change and the actual requirement of the limited change of fuel type, ensure that accuracy and the feasibility of biomass stove burning optimization.

Embodiment

A modeling method for biomass stove burning optimization, concrete steps are:

(1). gather the characteristic index of biomass stove operational factor and relevant characterising biological matter stove fired state, building database; Concrete biomass stove operational factor is obtained by biomass stove real-time running data storehouse, or directly measures collection by instrument and equipment.General biomass stove right biomass fuel kind (originating different with technical analysis data) limited, therefore, different biomass fuels (raw material of biomass fuel is different with source) is planted for given biomass stove burning n (n >=1), different types of fuel will separate image data, so that modeling respectively targetedly.

Described biomass stove operational parameter data comprises: the technical analysis index of biomass fuel, primary air velocity, secondary wind speed, oxygen amount, after-flame wind speed, biomass fuel delivery rate; The data of the characteristic index of described characterising biological matter stove fired state comprise NOx concentration in flue gas and biomass stove burning efficiency, and its method obtained is mature technology;

(2). the data in database are selected and pre-service, and use least square method supporting vector machine and radial base neural net (RBF network) for different fuel respectively, combustion model between the characteristic index setting up biomass stove operation parameter and fired state, therefore, be directed to n kind biomass fuel, set up n model respectively.Concrete grammar is:

Because biomass stove fuel change is limited, therefore respectively for the combustion case of different fuel, carry out data selection in a database, select the sample data of modeling, following principle is followed: be 1. evenly distributed during selection, namely it is uniform for spatially distributing at the topological structure of the input quantity of model, and the input quantity of selected data is not intensive concentrate on a point, can occupy a space uniformly; 2. equal in number, in topological structure, be namely positioned at the sample size of the data of difference, should be more or less the same in 10 ﹪ of the sample data amount of minimum data point;

Pre-service before modeling is carried out to the data selected, by the conversion of unit or the method for multiplying factor, makes the data of each input quantity and corresponding output quantity be in the order of magnitude and differ the state being less than 1, then input quantity is normalized;

The data that application pre-service is good, first least square method supporting vector machine algorithm is adopted, for a kind of fuel modeling, least square method supporting vector machine algorithm institute established model generalization ability is stronger, and then apply radial base neural net modeling, radial base neural net institute established model empiric risk is less, finally least square method supporting vector machine model and radial basis neural network is carried out being combined to form the final burning optimization model being directed to a kind of fuel; Output parameter for the input parameter of modeling and the characteristic index of characterising biological matter stove fired state is expressed as , wherein represent the organize the biomass stove operational factor vector as input data, represent the group as the parameter of the characterising biological matter stove fired state feature of output parameter, for sample size, based on actual operating data, set up the model between operation parameter fired state index different from biomass stove;

First, adopt the modeling of least square method supporting vector machine algorithm, kernel function elects radial basis function as:

for the width of radial basis function, this representation is canonical form; for mapping function, if required objective function is: , for the characteristic index predicted value of the boiler combustion status that model exports, for weight coefficient vector, for intercept; Introduce relaxation factor ξ ^* _iand ξ _iand allow error of fitting ε, ξ ^* _i>=0, ξ _i>=0, model is by retraining:

, under condition, minimize:

Obtain, wherein constant cfor penalty coefficient, c>0; This minimization problem is a convex quadratic programming problem, introduces Lagrangian function:

Wherein , , , for Lagrange's multiplier, >=0, >=0, >=0, >=0.

At saddle point place, function L is about w, b, ξ _i, ξ _i ^*, be also , , , maximal point, minimization problem is converted into the maximization problems asking its dual problem;

LagrangianL is about w, b, ξ at saddle point place _i, ξ _i ^*minimal point:

The dual function of Lagrangian function can be obtained:

Now,

According to Ku En-Plutarch (KKT) conditional theorem, following formula is had to set up at saddle point:

From above formula, α _i α _i ^* =0, α _i with α _i ^* can not be all non-zero simultaneously, can obtain:

Can b be obtained from above formula, obtain model;

Secondly, the radial base neural net modeling that learning ability sum functions approximation capability is strong is adopted:

For its output of the radial base neural net of individual hidden node is : , for weight coefficient, for dimension input vector, be the center of individual basis function, for the base width parameter of function; The key setting up radial basis neural network is the center determining basis function , sound stage width degree and weight coefficient ; Adopt particle cluster algorithm repetitive exercise radial base neural net, definition particle cluster algorithm initial population each dimension component of vector, be respectively hidden node, number Basis Function Center, the sound stage width degree of function and weight coefficient, objective function is: , wherein be the radial base neural net output valve of individual sample, be the actual value of individual sample; When reach minimum, when reaching setting value or complete iterations, training completes, obtain hidden node number, Basis Function Center, the sound stage width degree of function and weight coefficient, thus obtain radial basis neural network;

(3). determine the portfolio ratio of least square method supporting vector machine model and radial basis neural network; For the same biomass fuel that modeling data is corresponding, gather the data of new biomass stove under different running status as test samples, application least square method supporting vector machine model and the average weighted Forecasting Methodology of radial basis neural network, predict check data, namely , wherein be the target prediction value of group test samples operating mode, for least square method supporting vector machine model predication value, for radial basis neural network predicted value, for least square method supporting vector machine model predication value scale-up factor, for the prediction scale-up factor of radial basis neural network, and ;

with determination adopt the optimizing of ant group algorithm iteration to determine, initialization ant group position vector each dimension component, be respectively supporting vector machine model ratio with original Model Weight , objective function is: , wherein be the error of the biomass stove combustion characteristic index that group operating mode real data and combination model are predicted, K is test samples quantity, when achieve minimum, when reaching setting value or complete iterations, optimizing completes;

(4). least square method supporting vector machine model is combined with the scale-up factor of radial basis neural network by step (3) determined optimum, forms built-up pattern, namely , wherein Z is the built-up pattern after upgrading, thus realizes a kind of foundation of burning optimization model of biomass stove fuel;

(5). for other kind of biomass fuel modeling of given biomass stove; Modeling procedure is consistent with step (2) ~ (4), obtains the combustion model in different biomass fuel situation, total n kind fuel, therefore correspondence establishment n model as described in step (1);

(6). the burning optimization model group of different n kind biomass fuels is combined, forms a biomass stove burning optimization block mold; When needing calling model to predict, based on fuel index, calls corresponding burning optimization model and predicts.

Claims (1)

1. a modeling method for biomass stove burning optimization, is characterized in that the concrete steps of the method are:

Step (1). gather the characteristic index of biomass stove operational factor and relevant characterising biological matter stove fired state, building database;

The data of described biomass stove operational factor comprise technical analysis index, primary air velocity, secondary wind speed, oxygen amount, after-flame wind speed, the biomass fuel delivery rate of biomass fuel; Biomass stove operational factor is obtained by biomass stove runtime database, or is directly gathered by apparatus measures, and different biomass fuels separates image data, modeling respectively;

The data of the characteristic index of described characterising biological matter stove fired state comprise NOx concentration and the biomass stove burning efficiency of flue gas;

Step (2). the data in database are selected and pre-service, and use least square method supporting vector machine and radial base neural net for different fuel respectively, the combustion model between the characteristic index setting up biomass stove operational factor and fired state; Be directed to n kind biomass fuel, set up n model respectively, concrete grammar is:

Respectively for the combustion case of different fuel, carry out data selection in a database, select the sample data of modeling, follow following principle: be 1. evenly distributed during selection, it is uniform for namely spatially distributing at the topological structure of the input quantity of model; 2. equal in number, in topological structure, be namely positioned at the sample size of the data of difference, be more or less the same in 10 ﹪ of the sample data amount of minimum data point;

Pre-service before modeling is carried out to the data selected, by the conversion of unit or the method for multiplying factor, makes the data of each input quantity and corresponding output quantity be in the order of magnitude and differ the state being less than 1, then input quantity is normalized;

The data that application pre-service is good, first least square method supporting vector machine algorithm is adopted, for a kind of fuel modeling, least square method supporting vector machine algorithm institute established model generalization ability is stronger, and then apply radial base neural net modeling, radial base neural net institute established model empiric risk is less, finally least square method supporting vector machine model and radial basis neural network is carried out being combined to form the final burning optimization model being directed to a kind of fuel; Output parameter for the input parameter of modeling and the characteristic index of characterising biological matter stove fired state is expressed as wherein x _irepresent i-th group of biomass stove operational factor vector as input data, y _irepresent i-th group of parameter as the characteristic index of the characterising biological matter stove fired state of output parameter, N is sample size, the model between the characteristic index setting up operational factor fired state different from biomass stove based on actual operating data;

First, adopt the modeling of least square method supporting vector machine algorithm, kernel function elects radial basis function as:

$K(x_i,x_j)=\mathrm{\φ}\left(x_i\right)\·\mathrm{\φ}\left(x_j\right)=\exp \left|\left(\frac{{\left|\right|x_i-x_j\left|\right|}^2}{{2\mathrm{\σ}}^2}\right)\right|$

σ is the width of radial basis function, and this representation is canonical form; φ (x) is mapping function, if required objective function is: f (x _i)=w φ (x _i)+b, f (x _i) be the characteristic index predicted value of boiler combustion status that model exports, w is weight coefficient vector, and b is intercept; Introduce relaxation factor and ξ _iand allow error of fitting ε, ξ _i>=0, model is by retraining:

$\left\{\begin{array}{c}{y}_i-w\·\mathrm{\φ}\left(x_i\right)-b\≤\mathrm{\ϵ}+{\mathrm{\ξ}}_i\\ w\·\mathrm{\φ}\left(x_i\right)+b-y_i\≤\mathrm{\ϵ}+{\mathrm{\ξ}}_i^{*}\\ {\mathrm{\ξ}}_i\≥0\\ {\mathrm{\ξ}}_i^{*}\≥0\end{array},i=1,...,N,\right.$ Under condition, minimize:

$\min R(w,\mathrm{\ξ},{\mathrm{\ξ}}^{*})=\frac{1}{2}w\·w+c\underset{i=1}{\overset{k}{\mathrm{\Σ}}}\mathrm{\ξ}+{\mathrm{\ξ}}^{*}$

Obtain, wherein constant C is penalty coefficient, C>0; This minimization problem is a convex quadratic programming problem, introduces Lagrangian function:

$\begin{array}{c}L(w,b,\mathrm{\ξ},{\mathrm{\ξ}}^{*},\mathrm{\α},{\mathrm{\α}}^{*},\mathrm{\γ},{\mathrm{\γ}}^{*})=\frac{1}{2}w\·w+c\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(\mathrm{\ξ}+{\mathrm{\ξ}}^{*})-\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\α}}_i[y_i-({\mathrm{\ξ}}_i+\mathrm{\ϵ}+f\left(x_i\right))]\\ -\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\α}}_i^{*}[{\mathrm{\ξ}}_i^{*}+\mathrm{\ϵ}+f\left(x_i\right)-y_i]-\underset{i=1}{\overset{N}{\mathrm{\Σ}}}({\mathrm{\γ}}_i{\mathrm{\ξ}}_i+{\mathrm{\γ}}_i^{*}{\mathrm{\ξ}}_i^{*})\end{array}$

Wherein α _i, γ _i, for Lagrange's multiplier, α _i>=0, γ _i>=0,

At saddle point place, function L is about w, b, ξ _i, , be also α _i, γ _i, maximal point, minimization problem is converted into the maximization problems asking its dual problem;

LagrangianL is about w, b, ξ at saddle point place _i, minimal point:

$\left\{\begin{array}{c}\frac{\∂}{\∂w}L=0\→w=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}({\mathrm{\α}}_1-{\mathrm{\α}}_i^{*})\mathrm{\φ}\left(x_i\right)\\ \frac{\∂}{\∂b}L=0\→\underset{i=1}{\overset{N}{\mathrm{\Σ}}}({\mathrm{\α}}_i-{\mathrm{\α}}_i^{*})=0\\ \frac{\∂}{{\∂\mathrm{\ξ}}_i}L=0\→C-{\mathrm{\α}}_i-{\mathrm{\γ}}_i=0\\ \frac{\∂}{{\∂\mathrm{\ξ}}_i^{*}}L=0\→C-{\mathrm{\α}}_i^{*}-{\mathrm{\γ}}_i^{*}=0\end{array}\right.$

The dual function of Lagrangian function can be obtained:

Now,

$w=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}({\mathrm{\α}}_i-{\mathrm{\α}}_i^{*})\mathrm{\φ}\left(x_i\right)$

$f\left(x\right)=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}({\mathrm{\α}}_i-{\mathrm{\α}}_i^{*})K(x,x_i)+b$

According to Kuhn-Tucker condition theorem, following formula is had to set up at saddle point:

$\left\{\begin{array}{c}{\mathrm{\α}}_i[\mathrm{\ϵ}+{\mathrm{\ξ}}_i-y_i+f\left(x_i\right)]=0\\ {\mathrm{\α}}_i^{*}[\mathrm{\ϵ}+{\mathrm{\ξ}}_i+y_i-f\left(x_i\right)]=0\end{array}\right.i=1,...,N$

From above formula, α _iwith can not be all non-zero simultaneously, can obtain:

$\left\{\begin{array}{c}{\mathrm{\ξ}}_i{\mathrm{\γ}}_i=0\\ {\mathrm{\ξ}}_i^{*}{\mathrm{\γ}}_i^{*}=0\end{array}i=1,...,N\right.$

Can b be obtained from above formula, obtain model;

Secondly, the radial base neural net modeling that learning ability sum functions approximation capability is strong is adopted:

Its output of radial base neural net for n hidden node is y: w _ifor weight coefficient, x is that m ties up input vector, c _ibe the center of i-th basis function, ρ _ifor the base width parameter of function; The key setting up radial basis neural network is the center c determining basis function _i, sound stage width degree ρ _iand weight coefficient w _i; Adopt particle cluster algorithm repetitive exercise radial base neural net, each dimension component of definition particle cluster algorithm initial population Z-direction amount, be respectively hidden node number, Basis Function Center, the sound stage width degree of function and weight coefficient, objective function is: wherein be the radial base neural net output valve of i-th sample, it is the actual value of i-th sample; When J reach minimum, reach setting value or complete iterations time, training completes, and obtains hidden node number, Basis Function Center, the sound stage width degree of function and weight coefficient, thus obtains radial basis neural network;

Step (3). determine the portfolio ratio of least square method supporting vector machine model and radial basis neural network; For the same biomass fuel that modeling data is corresponding, gather the data of new biomass stove under different running status as test samples, application least square method supporting vector machine model and the average weighted Forecasting Methodology of radial basis neural network, predict check data, namely wherein be the target prediction value of i-th group of test samples operating mode, Z _zfor least square method supporting vector machine model predication value, Z _sfor radial basis neural network predicted value, α ' is least square method supporting vector machine model predication value scale-up factor, the prediction scale-up factor that β ' is radial basis neural network, and α '+β '=1;

α ' and the determination of β ' adopt the optimizing of ant group algorithm iteration to determine, each dimension component of initialization ant group position vector x, be respectively the prediction scale-up factor β ' of least square method supporting vector machine model predication value scale-up factor α ' and radial basis neural network, objective function is wherein ψ _ibe the error of the biomass stove combustion characteristic index that i-th group of operating mode real data and combination model are predicted, K is test samples quantity, when variance summation ψ achieve minimum, reach setting value or complete iterations time, optimizing completes;

Step (4). least square method supporting vector machine model is combined with the scale-up factor of radial basis neural network by step (3) determined optimum, forms built-up pattern, i.e. Z '=α ' Z _z+ β ' Z _s, wherein Z ' is the built-up pattern after upgrading, thus realizes a kind of foundation of burning optimization model of biomass stove fuel;

Step (5). for other kind of biomass fuel modeling of given biomass stove; Modeling procedure is consistent with step (2) ~ (4), obtains the combustion model in different biomass fuel situation, total n kind fuel, therefore correspondence establishment n model as described in step (1);

Step (6). the burning optimization model group of different n kind biomass fuels is combined, forms a biomass stove burning optimization block mold; When needing calling model to predict, based on fuel index, calls corresponding burning optimization model and predicts.